Fast calibration of a dynamic vegetation model with minimum observation data

The estimation and uncertainty analysis of parameters for dynamic vegetation models is a complex process. If one is mainly interested in parameter estimation, this can be done with simple global stochastic search methods, while uncertainty analysis is carried out with traditional first-order analysis, which significantly reduces the number of needed model evaluations. Within a nonlinear regression framework, where the misfit between model and observations is expressed as a sum of weighted squares, we model the dynamics of tropical forest with a size-structured Sinko-Streifer model and demonstrate the general calibration procedure on a virtual data set. A second case study on real data for a single species shows that surprisingly total stem number, basal area and aboveground biomass are the minimum observations needed for successful calibration. A third case study on real data for a three species group shows the prediction of successional states while only using the former reduced set of observations for calibration. The methodology is well suited for time consuming models, where only limited amount of forest site observations are available.